Welcome to the Infrared Image Noise Reduction Database in the Infrared Open Source Database
As the manufacturing technologies of infrared devices become mature, people more and more apply infrared imaging technologies to various fields of life. We have created a new infrared noise reduction data set, in which infrared image data in high and low quality are provided in pairs. We hope that our database can be a valuable resource for researchers to create better infrared image noise reduction models and test their performance, to improve technologies in the infrared image noise reduction field.
Thermal imagers convert infrared radiations to electrical signals using the detectors, and then the signals are amplified and converted to standard video signals to form infrared images displayed on the TV screen or monitor. Infrared images are obtained through "measuring" the radiated heat of objects, and therefore have the following characteristics compared with visible-light images: low resolution, weak contrast, low Signal-to-Noise Ratio (S/N), blur visual effect, little information, etc. Such images must be pre-processed using technologies such as image enhancement to have a relatively comfortable visual effect. In all imaging processes, noise can be produce due to external causes on infrared images. For example, the heat conduction effect and scattering in the air result in blur image edges and weak contrast, uneven temperature results in unevenly distributed speckle noises, and detector gains and internal electronic circuits can produce salt-and-pepper noise and stripe noise. These noises severely hamper visual effects and advanced function implementation in the future. In addition, unlike visible-light image data which can be produced directly through capturing and averaging multiple images, it is far more difficult to create an infrared image noise reduction data set. To fill this void, we have created an infrared image noise reduction database in which the data is in pairs. Real low-quality infrared image data is obtained through infrared imaging devices, and such images are processed in complicated procedures to obtain the corresponding high-quality infrared image data. In the database, there are 2000 image pairs (a total of 4000 images), based on infrared images in resolution of 256×192 with different noise levels captured in indoor and outdoor environments. Currently, such public real infrared image noise reduction data sets are rare, and the data set can provide sufficient and effective data for infrared image algorithm researches.
Our researches aim to create a real infrared noise reduction database, which resolves the issue that paired images cannot be directly obtained in image researches and provides paired high-quality and low-quality infrared imaging data for infrared image noise reduction algorithm researches. We have collected data from various kinds of infrared imaging devices, such as handheld infrared imagers and fixed infrared cameras. 14-bit infrared data in different periods, different scenarios, and different environments are captured and exported from devices. Note that the collected data is not the original infrared data but the digital images that are generated through photoelectric conversion with preliminary nonuniformity correction implemented in high- and low-temperature environments. However, stripe noise, speckle noise, and lens shadows are not removed, and the imaging quality is relatively low. Users can convert them into files in common image formats to view the imaging effect. Then we have processed the data with a series of complicated infrared noise reduction algorithms and obtained high-quality infrared image data which is in pairs with the original low-quality infrared image data. Two sets of data for you: 1) 8-bit data: The data is converted from 14-bit data through linear compression. That is, data with noise and data without noise are actually compressed to 8-bit data, and therefore the effective bit width is 8 bit; 2) 14-bit data: The data with noise is the 14-bit infrared data collected from infrared imaging devices. After the normalization and noise reduction algorithms are applied, the data is restored to 14-bit data to obtain the corresponding ground truth.
Example of reading data:
str ='001. dat' ;
col = 256;
row = 192;
fid = fopen(str, 'rb');
A = fread(fid，[col row]，'uint16') ;
#Loading nuc data
col = 256
row = 192
img = np.fromfile(path, dtype=np.uint16)
#Use the reshape function in the array of numpy to rearrange the data that has been read.
img = img.reshape(row, col)
Researchers in this research:
LIU Qing (email@example.com)
XU Zhaofei (firstname.lastname@example.org)
WANG Jiansheng (email@example.com)
WANG Shuigen (firstname.lastname@example.org)
-----------Copyright Statement Start------------
All rights reserved (c) 2021 IRay Technology Co., Ltd.
All rights reserved.
Hereby granted without any written permission, this permission contains no permission fee or permission loyalties. You can use, copy, modify, and distribute the data and documents for any legitimate purpose. However, you must comply with the following requirements: Its complete content must be in all copies of this database, and the original source of the database is IRay Technology Co., Ltd., which is clearly indicated in any publications of research results with this database used. IRay Technology Co., Ltd. does not bear responsibilities of any direct, indirect, special, accidental, or secondary damages caused by the usage of this database and documents in any circumstance, even though you have consulted IRay Technology Co., Ltd. before use. If you use this database, we will assume that you are familiar with this copyright statement.
IRay Technology Co., Ltd. does not make any promises, including but not limited to implied promises for marketability and applicability of special purposes. The database is provided "as is", and IRay has no obligations in forcible maintenance, support, update, enhancement, or modification.
-----------Copyright Statement End------------
We are offering the infrared security database scenario to the public free of charge. If you have used this database in researches, please thank our company for conducting researches:
IRay Technology Co., Ltd.
You can provide information here to submit the application form for this database (if your institution is a domestic university or college, please fill in the name of your institution as announced by the Ministry of Education of the People's Republic of China). After the application is approved, the link for downloading the database will be sent to your e-mail box.
If you have any questions when downloading, viewing, or using the data set, please contact the maintenance personnel of this website: GUAN Chunlei (E-mail address: email@example.com)